Daniel J. McDonald
Department of Statistics
University of British Columbia
16 July 2021
Outline:
Implemented with
Reproducible talk:
Outline:
Implemented with {evalcast} package + {covidcast} API
Reproducible talk: all code included
1. Sunday/Monday morning, make sure any data issues we know of are fixed
2. Flag and ‘correct’ other anomalies
Do these randomly so that we don’t smooth the signal too much.
Some data has other obvious issues …
Some data has other obvious issues …
3. Run the forecaster
n_locations)States: 0-1-2 weekly lags of deaths and 0-1-2 weekly lags of cases.
Counties:
## List of 3
## $ cases : num [1:10] 0 1 2 3 6 9 12 15 18 21
## $ fb-smoothed-hh-cli: num [1:4] 3 10 17 24
## $ dv-smoothed-cli : num [1:4] 3 10 17 24
4. Look at the results
5. The submission
Important lessons:
Out-of-sample evaluation (with proper as-of) is huge.
Modular “forecaster template” is really helpful.
Nonstationarity is hard.
On an equal footing, the best model beats the baseline by 20%. But, give the baseline 3 weeks of data, then it beats the best model by 20%.